[MPICH] rsh problem on Linux

[Syed Irfan]

Thanks for clarifying it.

I do the following

mpdboot -n 1 -r /usr/bin/rsh -v -f ~/mpd.hosts
which internally runs command mpd with options so rsh/ssh basically used 
to start the mpd's running on each of the hosts. After mpd is running on 
the hosts then mpi_spawn comm is via tcp/ip to mpd telling it to 
increase ring size.

So far as mpi application programs its comms uses (device=sock,ssm etc.)

I hope I am on the right track :)

Cheers

William Gropp wrote:
> At 09:46 PM 6/13/2005, Syed Irfan wrote:
>> Hi there,
>>
>>
>> Anthony Chan wrote:
>>> I strongly recommend you to use mpich2 instead of mpich-1.
>>> The latest mpich2 version 1.0.2 was released last friday, provides
>>> a different mechanism that does not rely on rsh/ssh.  You can get
>>> it in http://www.mcs.anl.gov/mpi/mpich2
>>> A.Chan
>>
>> Hmm mpich2-1.0.2 provides a different mechanism so as not to rely on 
>> rsh/ssh ? Having a look at the code for mpich2-1.0.1 pm/mpd seems to 
>> use  rsh/ssh? any thing changed in v1.0.2 ? that I am not aware of? Or 
>> are you talking about mpi program comms device settings?
> 
> The MPD process manager doesn't use ssh/rsh.  There is a convenience 
> script that helps launch the MPD process manager on different nodes; 
> this does use ssh/rsh to start the MPD demons and is probably what you 
> saw.  But the MPD process managers can be started without a remote shell 
> program.  Once the MPD system is started, ssh/rsh are not used; this is 
> very different from MPICH1.  In addition, it is relatively easy to 
> interface MPICH2 to other process managers; in fact, an MPICH2 
> executable can be started with either the MPD or the gforker PM without 
> change.
> 
> Bill
> 
> 
>> Cheers
>> Irfan
> 
> William Gropp
> http://www.mcs.anl.gov/~gropp